Analysis of unsteady fluid flows through flow imaging and neural implicit functions

• Grant number: 23/13431-1
• Support Opportunities: Scholarships abroad - Research Internship - Post-doctor
• Start date: January 01, 2024
• End date: August 31, 2024
• Field of knowledge: Engineering - Mechanical Engineering - Transport Phenomena
• Principal Investigator: William Roberto Wolf
• Grantee: Renato Fuzaro Miotto
• Supervisor: Mark Nelson Glauser
• Host Institution: Faculdade de Engenharia Mecânica (FEM). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: Syracuse University, United States
• Associated to the scholarship: 22/09196-4 - Improving the understanding of unsteady aerodynamic flows via high-fidelity simulations, analytical modeling and deep learning techniques, BP.PD

Abstract

The main objective of the present work is the application of machine learning and modal decomposition techniques to investigate unsteady flows from Shadowgraph, Schlieren and PIV images. The databases obtained during the doctorates of Dr. Miotto and Prof. Zigunov will be available for analysis, as well as experimental data from Prof. Glauser, and the high-fidelity simulations carried out by Prof. Yiyang Sun, also from SU, and by Prof. Datta Gaitonde's group, from The Ohio State University (OSU). One of the techniques that will be developed during the internship is the Doak decomposition, based on the momentum potential theory, using neural networks based on time-resolved Schlieren images. This decomposition allows the flow to be separated into thermal, hydrodynamic and acoustic components, therefore playing a key role in the analysis of compressible flows. To achieve this, a Poisson equation must be solved, which imposes barriers for its application in experimental settings where the boundary conditions are not well defined and/or the researcher does not have a proper Poisson solver. In this way, the resulting tool will provide an alternative to Doak's decomposition, and can be easily applied to the diagnosis of various flows without the need for expensive devices and without having to deal with the ambiguities of boundary conditions. Flow modal decomposition can then be applied to the specific hydrodynamic, acoustic or thermal fields obtained from Doak's formulation in order to observe specific coherent structures. (AU)